Merge branch 'link_detach'

Andrii Nakryiko says:

====================
This patch set adds new BPF link operation, LINK_DETACH, allowing processes
with BPF link FD to force-detach it from respective BPF hook, similarly how
BPF link is auto-detached when such BPF hook (e.g., cgroup, net_device, netns,
etc) is removed. This facility allows admin to forcefully undo BPF link
attachment, while process that created BPF link in the first place is left
intact.

Once force-detached, BPF link stays valid in the kernel as long as there is at
least one FD open against it. It goes into defunct state, just like
auto-detached BPF link.

bpftool also got `link detach` command to allow triggering this in
non-programmatic fashion.

v1->v2:
- improve error reporting in `bpftool link detach` (Song).
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>

include/linux/bpf.h

@@ -793,6 +793,7 @@ struct bpf_link {
 struct bpf_link_ops {
 	void (*release)(struct bpf_link *link);
 	void (*dealloc)(struct bpf_link *link);
+	int (*detach)(struct bpf_link *link);
 	int (*update_prog)(struct bpf_link *link, struct bpf_prog *new_prog,
 			   struct bpf_prog *old_prog);
 	void (*show_fdinfo)(const struct bpf_link *link, struct seq_file *seq);

include/uapi/linux/bpf.h

@@ -117,6 +117,7 @@ enum bpf_cmd {
 	BPF_LINK_GET_NEXT_ID,
 	BPF_ENABLE_STATS,
 	BPF_ITER_CREATE,
+	BPF_LINK_DETACH,
 };
 
 enum bpf_map_type {
@@ -634,6 +635,10 @@ union bpf_attr {
 		__u32		old_prog_fd;
 	} link_update;
 
+	struct {
+		__u32		link_fd;
+	} link_detach;
+
 	struct { /* struct used by BPF_ENABLE_STATS command */
 		__u32		type;
 	} enable_stats;

kernel/bpf/cgroup.c

@@ -814,6 +814,7 @@ static void bpf_cgroup_link_release(struct bpf_link *link)
 {
 	struct bpf_cgroup_link *cg_link =
 		container_of(link, struct bpf_cgroup_link, link);
+	struct cgroup *cg;
 
 	/* link might have been auto-detached by dying cgroup already,
 	 * in that case our work is done here
@@ -832,8 +833,12 @@ static void bpf_cgroup_link_release(struct bpf_link *link)
 	WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link,
 				    cg_link->type));
 
+	cg = cg_link->cgroup;
+	cg_link->cgroup = NULL;
+
 	mutex_unlock(&cgroup_mutex);
-	cgroup_put(cg_link->cgroup);
+
+	cgroup_put(cg);
 }
 
 static void bpf_cgroup_link_dealloc(struct bpf_link *link)
@@ -844,6 +849,13 @@ static void bpf_cgroup_link_dealloc(struct bpf_link *link)
 	kfree(cg_link);
 }
 
+static int bpf_cgroup_link_detach(struct bpf_link *link)
+{
+	bpf_cgroup_link_release(link);
+
+	return 0;
+}
+
 static void bpf_cgroup_link_show_fdinfo(const struct bpf_link *link,
 					struct seq_file *seq)
 {
@@ -883,6 +895,7 @@ static int bpf_cgroup_link_fill_link_info(const struct bpf_link *link,
 static const struct bpf_link_ops bpf_cgroup_link_lops = {
 	.release = bpf_cgroup_link_release,
 	.dealloc = bpf_cgroup_link_dealloc,
+	.detach = bpf_cgroup_link_detach,
 	.update_prog = cgroup_bpf_replace,
 	.show_fdinfo = bpf_cgroup_link_show_fdinfo,
 	.fill_link_info = bpf_cgroup_link_fill_link_info,

kernel/bpf/net_namespace.c

@@ -142,9 +142,16 @@ static void bpf_netns_link_release(struct bpf_link *link)
 	bpf_prog_array_free(old_array);
 
 out_unlock:
+	net_link->net = NULL;
 	mutex_unlock(&netns_bpf_mutex);
 }
 
+static int bpf_netns_link_detach(struct bpf_link *link)
+{
+	bpf_netns_link_release(link);
+	return 0;
+}
+
 static void bpf_netns_link_dealloc(struct bpf_link *link)
 {
 	struct bpf_netns_link *net_link =
@@ -228,6 +235,7 @@ static void bpf_netns_link_show_fdinfo(const struct bpf_link *link,
 static const struct bpf_link_ops bpf_netns_link_ops = {
 	.release = bpf_netns_link_release,
 	.dealloc = bpf_netns_link_dealloc,
+	.detach = bpf_netns_link_detach,
 	.update_prog = bpf_netns_link_update_prog,
 	.fill_link_info = bpf_netns_link_fill_info,
 	.show_fdinfo = bpf_netns_link_show_fdinfo,

kernel/bpf/syscall.c

@@ -3991,6 +3991,29 @@ static int link_update(union bpf_attr *attr)
 	return ret;
 }
 
+#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
+
+static int link_detach(union bpf_attr *attr)
+{
+	struct bpf_link *link;
+	int ret;
+
+	if (CHECK_ATTR(BPF_LINK_DETACH))
+		return -EINVAL;
+
+	link = bpf_link_get_from_fd(attr->link_detach.link_fd);
+	if (IS_ERR(link))
+		return PTR_ERR(link);
+
+	if (link->ops->detach)
+		ret = link->ops->detach(link);
+	else
+		ret = -EOPNOTSUPP;
+
+	bpf_link_put(link);
+	return ret;
+}
+
 static int bpf_link_inc_not_zero(struct bpf_link *link)
 {
 	return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? 0 : -ENOENT;
@@ -4240,6 +4263,9 @@ SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, siz
 	case BPF_ITER_CREATE:
 		err = bpf_iter_create(&attr);
 		break;
+	case BPF_LINK_DETACH:
+		err = link_detach(&attr);
+		break;
 	default:
 		err = -EINVAL;
 		break;

net/core/dev.c

@@ -8979,12 +8979,20 @@ static void bpf_xdp_link_release(struct bpf_link *link)
 	/* if racing with net_device's tear down, xdp_link->dev might be
 	 * already NULL, in which case link was already auto-detached
 	 */
-	if (xdp_link->dev)
+	if (xdp_link->dev) {
 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
+		xdp_link->dev = NULL;
+	}
 
 	rtnl_unlock();
 }
 
+static int bpf_xdp_link_detach(struct bpf_link *link)
+{
+	bpf_xdp_link_release(link);
+	return 0;
+}
+
 static void bpf_xdp_link_dealloc(struct bpf_link *link)
 {
 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
@@ -9066,6 +9074,7 @@ static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
 static const struct bpf_link_ops bpf_xdp_link_lops = {
 	.release = bpf_xdp_link_release,
 	.dealloc = bpf_xdp_link_dealloc,
+	.detach = bpf_xdp_link_detach,
 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
 	.fill_link_info = bpf_xdp_link_fill_link_info,
 	.update_prog = bpf_xdp_link_update,

tools/bpf/bpftool/Documentation/bpftool-link.rst

@@ -21,6 +21,7 @@ LINK COMMANDS
 
 |	**bpftool** **link { show | list }** [*LINK*]
 |	**bpftool** **link pin** *LINK* *FILE*
+|	**bpftool** **link detach *LINK*
 |	**bpftool** **link help**
 |
 |	*LINK* := { **id** *LINK_ID* | **pinned** *FILE* }
@@ -49,6 +50,13 @@ DESCRIPTION
 		  contain a dot character ('.'), which is reserved for future
 		  extensions of *bpffs*.
 
+	**bpftool link detach** *LINK*
+		  Force-detach link *LINK*. BPF link and its underlying BPF
+		  program will stay valid, but they will be detached from the
+		  respective BPF hook and BPF link will transition into
+		  a defunct state until last open file descriptor for that
+		  link is closed.
+
 	**bpftool link help**
 		  Print short help message.
 

tools/bpf/bpftool/bash-completion/bpftool

@@ -1122,7 +1122,7 @@ _bpftool()
             ;;
         link)
             case $command in
-                show|list|pin)
+                show|list|pin|detach)
                     case $prev in
                         id)
                             _bpftool_get_link_ids
@@ -1139,7 +1139,7 @@ _bpftool()
                     COMPREPLY=( $( compgen -W "$LINK_TYPE" -- "$cur" ) )
                     return 0
                     ;;
-                pin)
+                pin|detach)
                     if [[ $prev == "$command" ]]; then
                         COMPREPLY=( $( compgen -W "$LINK_TYPE" -- "$cur" ) )
                     else

tools/bpf/bpftool/link.c

@@ -22,6 +22,8 @@ static const char * const link_type_name[] = {
 
 static int link_parse_fd(int *argc, char ***argv)
 {
+	int fd;
+
 	if (is_prefix(**argv, "id")) {
 		unsigned int id;
 		char *endptr;
@@ -35,7 +37,10 @@ static int link_parse_fd(int *argc, char ***argv)
 		}
 		NEXT_ARGP();
 
-		return bpf_link_get_fd_by_id(id);
+		fd = bpf_link_get_fd_by_id(id);
+		if (fd < 0)
+			p_err("failed to get link with ID %d: %s", id, strerror(errno));
+		return fd;
 	} else if (is_prefix(**argv, "pinned")) {
 		char *path;
 
@@ -316,6 +321,34 @@ static int do_pin(int argc, char **argv)
 	return err;
 }
 
+static int do_detach(int argc, char **argv)
+{
+	int err, fd;
+
+	if (argc != 2) {
+		p_err("link specifier is invalid or missing\n");
+		return 1;
+	}
+
+	fd = link_parse_fd(&argc, &argv);
+	if (fd < 0)
+		return 1;
+
+	err = bpf_link_detach(fd);
+	if (err)
+		err = -errno;
+	close(fd);
+	if (err) {
+		p_err("failed link detach: %s", strerror(-err));
+		return 1;
+	}
+
+	if (json_output)
+		jsonw_null(json_wtr);
+
+	return 0;
+}
+
 static int do_help(int argc, char **argv)
 {
 	if (json_output) {
@@ -326,6 +359,7 @@ static int do_help(int argc, char **argv)
 	fprintf(stderr,
 		"Usage: %1$s %2$s { show | list }   [LINK]\n"
 		"       %1$s %2$s pin        LINK  FILE\n"
+		"       %1$s %2$s detach     LINK\n"
 		"       %1$s %2$s help\n"
 		"\n"
 		"       " HELP_SPEC_LINK "\n"
@@ -341,6 +375,7 @@ static const struct cmd cmds[] = {
 	{ "list",	do_show },
 	{ "help",	do_help },
 	{ "pin",	do_pin },
+	{ "detach",	do_detach },
 	{ 0 }
 };
 

tools/include/uapi/linux/bpf.h

@@ -117,6 +117,7 @@ enum bpf_cmd {
 	BPF_LINK_GET_NEXT_ID,
 	BPF_ENABLE_STATS,
 	BPF_ITER_CREATE,
+	BPF_LINK_DETACH,
 };
 
 enum bpf_map_type {
@@ -634,6 +635,10 @@ union bpf_attr {
 		__u32		old_prog_fd;
 	} link_update;
 
+	struct {
+		__u32		link_fd;
+	} link_detach;
+
 	struct { /* struct used by BPF_ENABLE_STATS command */
 		__u32		type;
 	} enable_stats;

tools/lib/bpf/bpf.c

@@ -603,6 +603,16 @@ int bpf_link_create(int prog_fd, int target_fd,
 	return sys_bpf(BPF_LINK_CREATE, &attr, sizeof(attr));
 }
 
+int bpf_link_detach(int link_fd)
+{
+	union bpf_attr attr;
+
+	memset(&attr, 0, sizeof(attr));
+	attr.link_detach.link_fd = link_fd;
+
+	return sys_bpf(BPF_LINK_DETACH, &attr, sizeof(attr));
+}
+
 int bpf_link_update(int link_fd, int new_prog_fd,
 		    const struct bpf_link_update_opts *opts)
 {

tools/lib/bpf/bpf.h

@@ -178,6 +178,8 @@ LIBBPF_API int bpf_link_create(int prog_fd, int target_fd,
 			       enum bpf_attach_type attach_type,
 			       const struct bpf_link_create_opts *opts);
 
+LIBBPF_API int bpf_link_detach(int link_fd);
+
 struct bpf_link_update_opts {
 	size_t sz; /* size of this struct for forward/backward compatibility */
 	__u32 flags;	   /* extra flags */

tools/lib/bpf/libbpf.c

@@ -7748,6 +7748,11 @@ struct bpf_link *bpf_link__open(const char *path)
 	return link;
 }
 
+int bpf_link__detach(struct bpf_link *link)
+{
+	return bpf_link_detach(link->fd) ? -errno : 0;
+}
+
 int bpf_link__pin(struct bpf_link *link, const char *path)
 {
 	int err;

tools/lib/bpf/libbpf.h

@@ -229,6 +229,7 @@ LIBBPF_API int bpf_link__unpin(struct bpf_link *link);
 LIBBPF_API int bpf_link__update_program(struct bpf_link *link,
 					struct bpf_program *prog);
 LIBBPF_API void bpf_link__disconnect(struct bpf_link *link);
+LIBBPF_API int bpf_link__detach(struct bpf_link *link);
 LIBBPF_API int bpf_link__destroy(struct bpf_link *link);
 
 LIBBPF_API struct bpf_link *

tools/lib/bpf/libbpf.map

@@ -273,6 +273,8 @@ LIBBPF_0.0.9 {
 
 LIBBPF_0.1.0 {
 	global:
+		bpf_link__detach;
+		bpf_link_detach;
 		bpf_map__ifindex;
 		bpf_map__key_size;
 		bpf_map__map_flags;

tools/testing/selftests/bpf/prog_tests/cgroup_link.c

@@ -2,6 +2,7 @@
 
 #include <test_progs.h>
 #include "cgroup_helpers.h"
+#include "testing_helpers.h"
 #include "test_cgroup_link.skel.h"
 
 static __u32 duration = 0;
@@ -37,7 +38,8 @@ void test_cgroup_link(void)
 	int last_cg = ARRAY_SIZE(cgs) - 1, cg_nr = ARRAY_SIZE(cgs);
 	DECLARE_LIBBPF_OPTS(bpf_link_update_opts, link_upd_opts);
 	struct bpf_link *links[ARRAY_SIZE(cgs)] = {}, *tmp_link;
-	__u32 prog_ids[ARRAY_SIZE(cgs)], prog_cnt = 0, attach_flags;
+	__u32 prog_ids[ARRAY_SIZE(cgs)], prog_cnt = 0, attach_flags, prog_id;
+	struct bpf_link_info info;
 	int i = 0, err, prog_fd;
 	bool detach_legacy = false;
 
@@ -219,6 +221,22 @@ void test_cgroup_link(void)
 	/* BPF programs should still get called */
 	ping_and_check(0, cg_nr);
 
+	prog_id = link_info_prog_id(links[0], &info);
+	CHECK(prog_id == 0, "link_info", "failed\n");
+	CHECK(info.cgroup.cgroup_id == 0, "cgroup_id", "unexpected %llu\n", info.cgroup.cgroup_id);
+
+	err = bpf_link__detach(links[0]);
+	if (CHECK(err, "link_detach", "failed %d\n", err))
+		goto cleanup;
+
+	/* cgroup_id should be zero in link_info */
+	prog_id = link_info_prog_id(links[0], &info);
+	CHECK(prog_id == 0, "link_info", "failed\n");
+	CHECK(info.cgroup.cgroup_id != 0, "cgroup_id", "unexpected %llu\n", info.cgroup.cgroup_id);
+
+	/* First BPF program shouldn't be called anymore */
+	ping_and_check(0, cg_nr - 1);
+
 	/* leave cgroup and remove them, don't detach programs */
 	cleanup_cgroup_environment();